雲端致能應用程式的動態資源規劃與目標效能管理

Abstract

隨著網路的的蓬勃發展，越來越多的應用程式必須透過網路來進行。此外，計算資源(CPU)的需求逐漸的從實體機器上轉嫁到網路中。然而，網路應用程式通常會因為當工作量增加而沒有給予額外的資源以供使用而造成效能衰退(performance degradation) 以及服務要求被阻擋(request blocking)。因此，在這篇論文中，我們側重於在虛擬化雲端環境中運行的網路應用程式的動態資源配置與目標效能的關係，以應付廣泛的工作量。在此論文中，我們特別專注於計算資源，因為它是唯一可以在虛擬雲環境動態配置的資源。 在為了保證目標效能（如響應時間(response time)），我們提出一個分析模型可以估算在給定工作量上所該分配的適當資源。通過使用這個模型，我們回答了一個重要的問題： 網路應用程式的效能與分配給的計算資源的關系。此外，我們也嘗試找出的應用程式的服務請求速率（或封包傳輸率）與dom0為了幫助虛擬機的處理網絡傳輸所消耗的計算資源的關係。通過我們的實驗，我們最後提供一個方法來估計應用程式的服務工作量需求，和一個直觀的線性模型來預測 dom0的為了幫助虛擬機的處理網絡傳輸所消耗的計算資源。

As computer network widely spread, more and more applications are provided through the Internet. Additionally, computing resource requirement is gradually transporting from local to the Internet. However, web applications usually suffer from performance degradation and request blocking when workload increases while no extra resources are allocated for the use. Hence, in this thesis, we focus on the relationship of dynamic resource provisioning and the target performance goals of web applications running on a virtualized cloud environment so to cope with a wide varying workload. In particular, we focus on the computing resource since it is the only resource that can be dynamically configured in virtualized cloud environment. In order to guarantee the performance (i.e. response time), we propose an analytic model which can estimate proper resource allocation based on wide range workload. By using this model, we answer an important question which is how does a web server’s performance depend on the amount of CPU. Furthermore, we also try to discover the relationship of request rate (or packet rate) and the CPU utilization of Dom0 due to the network IO of the VM. Through our experiments, we provide a method to estimated service demand, and an intuitive linear model to predict the overhead of dom0.